Time, space, and episodicity of physical disturbance in streams
Daniel Millera,*, Charlie Luceb, Lee Bendac
aEarth Systems Institute, WA 3040 NW 57th Street, Seattle, WA 98107, USAbBoise Aquatic Sciences Lab, Rocky Mountain Research Station 316 E. Myrtle, Boise, ID 83702, USA
cEarth Systems Institute, CA 310 N Mt. Shasta Blvd., Suite 6, Mt. Shasta, CA 96067, USA
Abstract
Storm-driven episodes of gully erosion and landsliding produce large influxes of sediment to stream channels that have both
immediate, often detrimental, impacts on aquatic communities and long-term consequences that are essential in the creation and
maintenance of certain channel and riparian landforms. Together, these effects form an important component of river
ecosystems. In this paper, we describe issues involved in characterizing and predicting the frequency, magnitude, spatial
extent, and synchrony of these sediment influxes. The processes that drive sediment fluxes exhibit spatial and temporal
variability over a large range of scales. Disregard of this variability can have unanticipated consequences for efforts to quantify
process rates, as we illustrate using landslide densities observed for a storm event in western Oregon. Multiple factors interact to
create the temporal and spatial patterns of erosional and mass-wasting events that affect stream channels. Fires, in particular,
enhance susceptibility to erosional and mass-wasting processes, and thus affect the timing and magnitude of sediment-
mobilizing events. We use examples from west-central Idaho to show how fires, storms, and topography interact to create
spatially distinct patches of intense erosional activity. We require quantitative descriptions of these controlling factors to make
quantitative predictions of how differences or changes in topography, fire regime, and climate will affect the regime of sediment
fluxes. The stochastic and heterogeneous nature of these factors leads us to quantify them in probabilistic terms. The effects of
future fire and storm sequences are governed in part by the past sequence of events over time frames spanning centuries and
spatial extents spanning entire river basins. Empirical characterization of past events poses a considerable challenge, given that
our observational record typically spans several decades at most. Numerical models that simulate multiple event sequences
provide an alternative means for estimating the influence of antecedent conditions and for quantifying the role of different
controlling factors.
# 2003 Elsevier Science B.V. All rights reserved.
Keywords: Aquatic habitat; Fire; Landslides; Erosion
1. Introduction
Intense or extended rainfall can trigger surface
erosion and landsliding (Caine, 1980) that bring large
and sudden pulses of sediment and organic debris to
stream channels. These events can substantially
impact channel and riparian habitats. Landslides, deb-
ris flows, and consequent debris torrents can scour
channels to bedrock and destroy riparian vegetation
(Hack and Goodlett, 1960; Benda, 1990; Nolan and
Marron, 1990; Cenderelli and Kite, 1998; May, 1998);
local deposition from landslides and debris flows can
bury or block channels (Nolan and Marron, 1988),
create log jams (Hogan et al., 1998), and potentially
Forest Ecology and Management 178 (2003) 121–140
* Corresponding author. Tel.: þ1-206-633-1792;
fax: þ1-425-671-0094.
E-mail address: [email protected] (D. Miller).
0378-1127/03/$ – see front matter # 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0378-1127(03)00057-4
create conditions conducive to dam-break floods
(Coho and Burges, 1993); sediment introduced by
mass wasting and extensive gully erosion can alter
channel characteristics both locally and over many
kilometers, with effects that include channel widen-
ing, reductions in pool frequency, fining of bed
texture, and increased turbidity (Coates and Collins,
1984; Everest et al., 1987; Nolan and Marron, 1990;
Harvey, 1991; Madej and Ozaki, 1996; Montgomery
and Buffington, 1998).
Although these impacts can be detrimental to aqua-
tic communities, the erosional and mass wasting
events that trigger them are also recognized as integral
to the creation and maintenance of certain types of
channel habitat. Landslides and debris flows bring
boulders and large woody debris that provide long-
term sources of channel roughness and complexity
(Everest and Meehan, 1981; Benda, 1990; Grant et al.,
1990; Wohl and Pearthree, 1991; Grant and Swanson,
1995), large, transient increases in sediment supply
create berms, terraces, and fans that shape the valley
floor (Benda, 1990; Madej, 1990; Nakamura et al.,
1995) and create side channels and riparian surfaces
(Miller and Benda, 2000).
The impacts of erosional and mass-wasting events
evolve over time as fluvial transport moves and re-
sorts channel-stored sediment, debris jams decay,
riparian vegetation regrows, and large woody debris
is recruited to channels (Benda, 1990; Nolan and
Marron, 1990; Grant and Swanson, 1995; Hogan
et al., 1998; May, 2001; Pabst and Spies, 2001).
Conditions encountered in channels subject to pulses
of sediment input and transport thus depend on where
in time one intersects this trajectory of change. The
population of sub-basins and tributary channels that
constitute a drainage basin provides many separate
and potentially independent sediment sources. Asyn-
chronous activation of these sources produces a popu-
lation of channel reaches at different points along the
post-event trajectory of channel evolution (Benda
et al., 1998). The past sequence of sediment influx
and transport events thus acts to create and enhance
spatial heterogeneity in channel conditions (Benda
et al., this issue) and contributes to habitat complexity.
Regionally, the mosaic of habitat conditions created
by the sequence and spatial distribution of these events
forms an important component of riverine ecosystem
structure (Reeves et al., 1995).
Fires play an integral role in the timing and severity
of erosional and mass-wasting events (Swanson, 1981;
Meyer et al., 1995; Cannon, 2001; Istanbulluoglu et al.,
2002). Fires can destroy ground cover, reduce soil
infiltration capacity, and kill vegetation, all of which
increase the potential for rainfall-triggered erosion
and mass wasting (Wondzell and King, this issue).
These effects are greatest immediately post-fire and
tend to dissipate over several years, although fire-
related loss of root strength and the associated sus-
ceptibility to landsliding may persist for a decade or
more (Meyer et al., 2001). Thus fires and storms act
together to drive sediment fluxes over a landscape. The
sequence of sediment inflow and transport events
occurring within a river basin arises from the inter-
acting sequence of fires and storms (Benda et al.,
1998). The impacts to stream channels vary over time,
depending on where they are within this sequence.
Other processes, such as disease, windthrow, and snow
avalanches, can also alter vegetation cover and affect
susceptibility to surface erosion and mass wasting, but
fire is the predominant source of vegetation distur-
bance in many landscapes (Agee, 1993).
Storm-driven surface erosion and mass wasting,
accentuated by fire, generate large inputs of sediment
and organic debris to channel systems that are punc-
tuated in space and time (Benda and Dunne, 1997b).
These inputs act to alter the suite of channel and
riparian habitat types and quality found in a basin,
adding to the dynamic and heterogeneous elements of
the river environment. These elements have character-
istic temporal and spatial scales that are governed by
the frequency and magnitude of the sediment inflows,
which in turn are governed by the periodicity, magni-
tude, and timing of fires and storms, modulated by
rates of sediment production (Benda and Dunne,
1997a).
The regime of sediment inflows can be difficult to
characterize, in part because of the extended periods
that may separate large-magnitude events (Kirchner
et al., 2001). Yet, as discussed above, this regime
forms an integral component of a river ecosystem
(Reeves et al., 1995; Gresswell, 1999), and any
changes to that regime pose consequences that are
largely unknown. To anticipate the effects of natural
and anthropogenic change in fire and climate on
stream environments, we must characterize the effects
those changes have on the frequency and magnitude of
122 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
sediment influxes and the impact of those influxes on
channels. Benda et al. (this volume) approach the
second topic; here we address the first.
Unfortunately, data to directly quantify the controls
on frequency and magnitude of sediment fluxes are
limited, in large part because of the long periods
involved, so that our understanding of these processes
is largely qualitative and based on inference. Meyer
and Pierce (this volume), show that erosional regimes
have changed over time concurrent with climatic
shifts, but we do not yet have quantitative models
to predict how current and future changes in climate
and fire regime will alter the timing and magnitude of
erosion events. Indeed, it is difficult to quantitatively
characterize current erosion regimes. Erosion and
mass-wasting processes are unevenly distributed in
time and space, which creates a particular challenge
for empirical characterization, as we illustrate using
measures of landslide density in coastal Oregon. Multi-
ple interacting processes drive erosion and mass-wast-
ing events, which complicate attempts to characterize
the factors controlling their occurrence. As an example,
we look at how patterns of stream disturbance depend
on the timing and spatial coincidence of fires and
storms and at how topography and storm characteris-
tics interact to control the spatial distribution of land-
slides in west-central Idaho. Finally, we briefly discuss
methods for quantifying these processes and describe
the use of numerical models for characterizing the
controls on sediment flux.
2. Scale relations and sources of variability
Landslides and debris flows are an important
mechanism for storm-driven pulses of sediment
delivery to channels in many landscapes (Hack and
Goodlett, 1960; Larsen and Simon, 1993; Benda and
Dunne, 1997b; Hovius et al., 1997) and characteriza-
tion of landslide rates is a key component in efforts
to quantify effects of land management on erosion
regimes (Sidle et al., 1985; Montgomery et al., 2000;
Brardinoni, 2001). In February 1996, a high-intensity,
long-duration storm triggered landsliding and flood-
ing across western Oregon. Rainfall intensities and
flood peaks exhibited great spatial variability (Taylor,
1997a); estimated return intervals for flood peaks in
the area spanned a range from less than 10 to over
100 years (Bush et al., 1997). Several studies examined
the effect of forest cover and land management on
landslide densities associated with this storm (Bush
et al., 1997; May, 1998, 2002; Robison et al., 1999).
Although these studies did not include fire-related
landslides, the results are still instructive.
These three studies reported a large range in
observed landslide densities, arising in part from large
spatial variability in storm intensity and from differ-
ences in the study methods. Bush et al. (1997) relied
on landslides identified on 1:24,000-scale aerial
photographs, whereas Robison et al. (1999) and
May (1998, 2002) used field surveys to locate land-
slides and debris flows that reached stream channels.
The study of Bush et al. (1997) encompassed an area
exceeding 4000 km2 including the Siuslaw National
Forest (SNF) and surrounding areas; the Robison et al.
(1999) study involved six sites in western Oregon
following the February storm encompassing an area
of almost 83 km2; May’s survey sites involved 11
third- to fifth-order streams in the Siuslaw Basin with
a combined area of 47 km2. Each survey resulted in
different relationships between relative landslide den-
sity and forest-stand type. All studies found the high-
est average densities in clearcut areas harvested within
the last decade. Results differed substantially, how-
ever, for the relative density of landsliding between
second-growth and older stands. Robison et al. (1999)
consistently found higher densities in stands older than
100 years, whereas May (2002) consistently found
higher densities in second-growth stands (Table 2,
p. 1103). Bush et al. (1997) did not differentiate
between second-growth and older stands. These dif-
ferences illustrate the difficulty in characterizing a
heterogeneous entity and provide an opportunity to
look at how the scale of measurement can affect such
observations.
The Coastal Landscape and Analysis Study
(CLAMS; http://www.fsl.orst.edu/clams/) used data
from Bush et al. (1997) and Robison et al. (1999)
to derive regionally applicable estimates of landslide
susceptibility as functions of forest-cover type for the
Coast Range of Oregon. To examine the effects of
sampling scale, the Bush et al. (1997) inventory
(Fig. 1) was subsampled with replacement (a bootstrap
sample, e.g. Chernick, 1999) at a variety of scales
with landslide densities calculated as a function of
forest type for each sample. Forest type was evaluated
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 123
using the CLAMS vegetation coverage (Ohmann and
Gregory, 2002), with forest classes grouped into three
broad categories: (1) OPEN, consisting of recently
clearcut and unforested land, (2) MIXED, consisting
of forest cover of predominately small conifers or
forests with a predominance of hardwoods, and (3)
LARGE, forest cover consisting predominately of
large conifers. A fourth class for roads was also
included in the CLAMS study, but is not used for
our analysis. Data from Robison et al. (1999) were
used to estimate bias in density estimates arising from
the inability to see small landslides under tree canopy
in aerial photographs (Pyles and Froehlich, 1987;
Brardinoni, 2001) and to account for spatial variability
in topographic control on landslide susceptibility
(Miller et al., 2002).
Fig. 1. Landslide-inventory sites used for evaluating scale effects on landslide density. The SNF study (Bush et al., 1997) encompassed over
4000 km2 in two blocks, indicated by the green polygons on the map of western Oregon. The yellow polygons indicate the study sites included
in the ODF study (Robison et al., 1999). Landslide locations mapped in the SNF study are shown with white dots on the shaded relief images
to the right. Forest-cover types described in the text are indicated by color.
124 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
We found that the relative landslide density between
these three cover types varied as a function of scale.
For sample areas spanning tens of square kilometers,
there is a high probability that observed landslide
density is highest in the LARGE cover type (Fig. 2),
as observed by Robison et al. (1999). But at scales of
hundreds of square kilometers, observed densities are
almost always lowest in the LARGE cover type.
This dependence on sample scale arises in part
because the sample area is smaller than the character-
istic scale of heterogeneity in the spatial distribution of
landslides, a concept related to the representative
length in continuum mechanics (e.g. Middleton and
Wilcock, 1994) and the elementary unit volume in
fluid dynamics (Bear, 1972). We found that the range
of observed landslide densities varied as a function of
the area sampled (Fig. 3). A large range in densities is
found for small sample areas, with a large proportion
of the samples containing no landslides at all. Two
changes occur as the sample area increases: excluding
zero values, both the range and modal value of
observed landslide densities asymptotically decrease.
The decrease in modal value (Fig. 3) occurs because
about 20% of the samples—those with zero values,
primarily small sample areas—are excluded. Similar
biasing occurs with field studies, because site surveys
and resulting conclusions focus on areas that have
landslides. With all samples included, the modal value
remains constant with sample area. If the sample area
is small, there is a probability of observing a high
landslide density. If the area covered by one particular
cover type in a sample tends to be smaller than the
others, there is a probability of finding a higher land-
slide density in that cover type solely because of the
differences in area.
Excluding roads, the Bush et al. (1997) study area
included 21% in the OPEN class, 55% in the MIXED
class, and 24% in the LARGE class. Robison et al.’s
(1999) six sites included 17% in age classes less than
10 years, 61% in age classes from 10 to 100 years, and
22% in age classes exceeding 100 years. May’s (1998,
2002) sites included 23% in clearcut units, 47% in
second-growth stands, and 30% in mature forests. This
difference in area between cover types may account
for the behavior observed for the MIXED and LARGE
classes (Fig. 2), because most samples have a smaller
proportion of their area in the LARGE class. It does
not explain the relationship between the OPEN and
LARGE classes, however, since these two classes
involve similar areas. This suggests that the length
scale required to average variability in the LARGE
cover class is greater than for the MIXED class,
Fig. 2. The landslide inventory from the SNF study (Fig. 1) was subsampled over a range of scales. For any contiguous sample block
containing landslides in more than one cover type, the probability that landslide density in the ‘‘LARGE’’ cover type is greater than that in
either the ‘‘OPEN’’ or ‘‘MIXED’’ cover types varies as a function of the size of the sample block.
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 125
although other unexplored sampling issues may be
involved.
These results show that landslides are found in
dense clusters over tens of square kilometers with
large intervening areas containing few landslides.
They also suggest that landslides tend to be clustered
more tightly in old forests and more evenly distributed
in other forest types. Conclusions regarding the
influence of vegetation on landslide susceptibility
may thus vary dramatically depending on the scale
of observation. The large scale of heterogeneity that
must be accommodated complicates characterization
of landslides and other sediment-moving events.
The scale effects described above result from varia-
bility in the attributes being measured, which in this
case are landslide locations. Many factors create
variability in the production and delivery of sediment
to stream channels. Observed differences in the spatial
density of landslides are explained in part by varia-
tions in geology, topography, and vegetation (e.g.
Dragovich et al., 1993a,b). Such controls can be
quantified and, and to the extent and resolution that
they are mapped, their relative effect on landslide rate
can be anticipated, as is done for landslide-hazard
mapping (e.g. Hammond et al., 1992).
There are other factors whose influences on land-
slide location are unknown a priori. Spatial variability
in storm characteristics provides an example. The
storm in February 1996 that triggered the landslides
mapped in Oregon, although regionally extensive,
exhibited large variability in rainfall intensity (Oregan
State University, 1996; Taylor, 1997b, see also http://
www.ocs.orst.edu/gifs/flood_map.GIF). Convective
storms can also generate local high-intensity cells over
spatial scales of kilometers. Spatial variability in storm
intensity thus creates heterogeneity in sediment pro-
duction rates over scales spanning distances from one
to hundreds of kilometers.
The distribution of soil depths poses another
unquantified source of variability in factors controlling
sediment production (Dunne, 1998; Schmidt, 1999).
Soil depth is an essential factor in stability of hillslope
soils (Hall et al., 1994) and in setting the volume of
sediment available for delivery to a channel system
(Benda and Cundy, 1990). The spatial distribution of
soil depth over any area, coupled with the future
sequence of fires and storms, determines in part the
number of landslides that will occur within that area
over any specified period of time. If a series of storms
triggers a large number of landslides that evacuate the
soil from landslide-prone hollows, the distribution of
soil depths is changed along with the potential for
future landsliding. The spatial distribution of soil depth
and the potential for sediment delivery to the channel
system is thus a function of past events (Benda and
Dunne, 1997b). Given the time scales involved for
colluvial refilling of hollows (Dietrich and Dunne,
1978; Reneau et al., 1990), that history may span many
centuries. The past sequence of storms, fires, and other
vegetation disturbances, such as disease and wind-
throw, create a mosaic of soil properties with variability
over scales spanning meters to hundreds of kilometers.
Our recourse in dealing with effects of stochastic
and heterogeneous processes like storms and fires is to
Fig. 3. The upper graph is a scatter plot of landslide densities from
50,000 randomly placed sample blocks within the two SNF study
areas shown in Fig. 1. Densities are shown for the entire sample
block with no separation of cover type. The slanted arrays of
aligned dots correspond to samples containing one landslide (the
lowest, left-most array), two landslides, and so on. About 20% of
the samples had no landslides. The proportion of samples with no
landslides is shown as a function of sample size in the lower graph.
126 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
characterize them empirically in terms of probability
distributions. The use of probability distributions is
well established for estimating storm intensities and
peak flows for specific intervals of time. We can do the
same with patterns of spatial heterogeneity in sedi-
ment production arising from unquantified sources,
such as potential variations in soil depth. For example,
soil geotechnical properties are variable over scales
affecting individual landslides and gullies and prob-
abilistic descriptions of soil properties are used in
estimates of slope stability (Hammond et al., 1992).
2.1. Stochastic interactions: temporal–spatial
correlations
Stochastic erosional drivers acting over a heteroge-
neous topography create an erosion regime character-
ized by episodic patches of activity. The size and
location of these patches depends on the interactions
of fires, storms, and topography, which we illustrate
here with examples from central Idaho. The Boise
National Forest has experienced numerous large fires
over the last century (Fig. 4). During the last 12 years,
fire intensity, measured by crown scorch, was mapped
from the air for several of these fires. Over this time
period, landslides, gully erosion, and evidence of stream
disturbance were also mapped from 1:15,800-scale
aerial photographs (Fig. 5). Mapped patch sizes for
variations in fire intensity have length scales on the
order of 1–5 km. These fires had a decidedly non-uni-
form influence on patterns of mass wasting and asso-
ciated channel disturbance. The ‘‘patches’’ of stream
disturbance are individually larger than the patches of
high-intensity burn; stream disturbances may initiate
within high-intensity burn patches and then propagate
through lower-intensity burn and unburned areas.
Disturbed channels concentrated in patches with
length scales on the order of 5–10 km are also evident
(Fig. 5). These patches cluster in and near recent fires,
but several fall outside burned areas. Examination of
event timing reveals a mixture of causes. At large
(between group) scales, much variability is explained
by whether or not a severe weather event struck the
area. One particularly large thunderstorm in summer
1996 (prior to the aerial photography) affected several
streams just slightly north and west of the center of the
basin. The cluster in the southwestern part of the basin
was affected primarily by a thunderstorm in 1993,
shortly after the fire. Although not shown on this map,
the 1 January, 1997 rain-on-snow event affected
patches in the southwestern part of the basin again,
and several streams immediately north of the mapped
area in an area burned in 1989. The clustering in these
groups is similar to that seen in Fig. 5. All affected
streams were below 1500 m in elevation. Variations in
burn intensity explain only part of the controls on mass
wasting in individual basins in the area of that storm.
The debris flow on the far-eastern side of the basin
resulted from a small thunderstorm centered over one
flank of the basin. After the event, one could walk
across a hillslope with relatively uniform burn severity
and see the erosional response change from severely
eroded to undisturbed within 500 m although no var-
iation in fire severity (as measured by soil character-
istics) was observed over that transect. These
observations suggest that the scale of variability in
the driving weather exerts a strong control on the
spatial extent of subsequent erosion.
It is useful to compare the patch size of stream
disturbance seen in these examples to the size of
habitat patches used by fish (Fig. 6). Variations in
summer water temperature are the hypothesized
source of fragmentation for bull trout (Salvelinus
confluentus) in this basin, which yields elevation as
a control on the distribution of spawning and rearing
habitat (Dunham and Rieman, 1999). Dunham and
Rieman (1999) found that larger patches were more
likely to be occupied, in part because larger patches
yield larger, more diverse, more stable, and better-
connected populations of fish. Smaller patches are also
at risk of losing all of their habitat to mass wasting or
channel disturbance during a single event, whereas a
coherent set of disturbances over a patch the size of the
larger basins has not been seen in the last 15 years,
although fires of a size large enough to cover several
habitat patches are not unusual in the historical record.
If the spatial extent of stream disturbances are con-
trolled more by the intersection of fire and weather
than by the occurrence of fire alone, then fire alone is
unlikely to cause extirpation of more than one or a few
local populations during a given fire episode.
Interactions between climatic events and topogra-
phy also play an important role in setting the patch
scale over which erosional events are concentrated.
For example, the distribution of landslides mapped by
the Payette National Forest following a rain-on-snow
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 127
event in January 1997 revealed that elevation was
influential (Fig. 7). Most (94%) of the landslides were
between elevations of 1000 and 1500 m. Focusing
on this elevation range, 87% of these landslides
occurred on slopes steeper than 278 (estimated from
a DEM). Here we see a relatively uniform storm, a
large synoptic system with high spatial correlation in
precipitation, coming over complex topography to
form a distinct pattern in geomorphic response.
2.2. Controls on event occurrence
To quantify relationships of cause and effect on
the frequency and magnitude of stream disturbance
Fig. 4. Fires over 100 ha mapped since 1908 on the Boise National Forest. Some areas have burned up to five times during that period.
128 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
caused by sediment-moving events, it is useful to
define a conceptual model of the processes involved.
From the discussion above, and capitalizing on work
by Benda and Dunne (1997a,b; further described in
Benda et al., 1998), we identified three primary com-
ponents:
(1) A spatial template imposed by topography,
bedrock lithology, geologic structure, and soil
type that controls locations of sediment produc-
tion and storage and sets points of delivery to the
channel system.
(2) A set of stochastic temporal drivers that alter
erosional susceptibility and trigger sediment
fluxes, e.g. fires and storms.
(3) The antecedent sequence of events, which with
the rate of sediment production, determines the
volume of sediment available for delivery to and
transport through the channel system.
These three concepts provide a framework with
which to interpret and anticipate differences in dis-
turbance regimes, but to be useful in quantifying any
differences, we must develop quantitative character-
izations of each component and of interactions
between them. We describe strategies for charac-
terizations of topography, fire regime, and storm
climate in the following sections. Characterization
of the antecedent sequence of events, or at least of
their consequences, poses a considerable challenge.
Fig. 5. Middle Fork Boise River basin showing fires since 1989 and streams with evidence of major bed disturbance. Estimates of fire intensity
are based on extent of crown scorch and disturbed channels are mapped from aerial photographs.
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 129
For example, we have no means for efficient high-
resolution measurement of soil depths over watershed
scales. Likewise, the large temporal and spatial scales
involved hinder quantification of process interactions.
In the final section, we discuss use of numerical models
as a means of addressing these issues and as a tool for
exploring the role of different factors in controlling the
frequency and magnitude of sediment fluxes.
2.2.1. Topography
Topographic attributes represent large controls on
susceptibility to mass wasting (e.g. Dietrich et al., 2001)
and surface erosion (Dietrich et al., 1992). Topography
also determines where in the landscape colluvium will
accumulate (Hack and Goodlett, 1960; Dietrich and
Dunne, 1978) and dictates the points of delivery for
water-eroded and mass-wasting derived sediment to
the channel network (Swanson et al., 1988; Benda
and Dunne, 1997b). The advent of digital elevation
data and a variety of tools and algorithms for using
it (e.g. Zevenbergen and Thorne, 1987) make spa-
tially distributed estimates of topographic attributes
straightforward. Coupled with field observations of
erosional processes, topographic data can be used to
Fig. 6. Channels with major channel disturbance events since 1989 in the Middle Fork Boise River basin juxtaposed with bull trout habitat
patches (after Dunham and Rieman, 1999).
130 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
constrain the probability of erosional inputs to channel
systems (Istanbulluoglu et al., 2002).
We illustrate these concepts using an example for
debris flows in the Oregon Coast Range, based on
work done with Coastal Landscape and Analysis
Study (http://www.fsl.orst.edu/clams/). Although this
example deals with a particular mass-wasting process,
the techniques are applicable to any topographically
controlled erosional process.
It is convenient to separate topographic controls on
debris-flow occurrence between those that affect
initiation and those that affect the downslope scour,
transport, and deposition of material. Debris flows are
often initiated by shallow landsliding of colluvial
Fig. 7. Landslides following January 1997 rain-on-snow event within study areas in the Payette National Forest. Lighter shading indicates
areas between 1000 and 1700 m elevation.
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 131
material accumulated in topographic hollows (Dietrich
and Dunne, 1978). Stability of shallow colluvial
deposits can be estimated using a simple model—
the infinite slope approximation—with a topographic
dependence on slope gradient alone (e.g. Hammond
et al., 1992). Shallow landslides are triggered by
increased pore-pressure gradients during periods of
intense precipitation (Caine, 1980), which can be
estimated as a function of rainfall intensity, upslope
contributing area, and local slope (O’Laughlin, 1986).
These two models point to local surface gradient and
specific contributing area as primary topographic
controls on shallow landsliding (Montgomery and
Dietrich, 1994). Iverson (2000) highlights limitations
to the hydrologic assumptions used in this model, but
still recognizes the importance of local slope and
specific contributing area in setting the antecedent
soil moisture conditions that modulate the impact of
transient periods of high-intensity rainfall on slope
stability. We thus use a function of local slope and
specific contributing area (that of Montgomery and
Dietrich (1994), with soil parameters held uniform), as
a topographic index of slope stability. Using landslide
inventories, this index can then be calibrated as a
function of relative landslide density. This provides
a spatially distributed estimate of topographic control
on susceptibility to shallow landsliding and associated
debris-flow initiation.
Once initiated, subsequent downslope travel dis-
tance of a debris flow can be parameterized in terms of
several topographic attributes. Benda and Cundy
(1990) identified channel gradient and channel junc-
tion angles along the debris-flow track as dominant
controls. CLAMS (Miller et al., 2002) has expanded
on their work to include cumulative scour and de-
positional length as a proxy for debris-flow volume
(Iverson et al., 1998; May, 2002) in a probabilistic
model of debris-flow runout. This model was calibrated
using field mapping done by the Oregon Department of
Forestry (ODF) following the 1996 storms (Robison
et al., 1999) in the Oregon Coast Range.
Coupling probability estimates of landslide initia-
tion and debris-flow runout provides an estimate for
the probability of debris-flow delivery of material to a
channel (Fig. 8). For example, in the Siuslaw River
basin in coastal Oregon, topographic differences
between Knowles Creek basin, to the east, and Sweet
Creek basin, to the west, result in large differences in
the predicted probability of debris-flow delivery
between these two channel systems. Large variability
exists even within smaller sub-basins within each sys-
tem. We expect that such differences in the topographic
controls on debris-flow behavior between basins will
affect the frequency and magnitude of sediment deliv-
ery to the channel system, with consequences for the
frequency of habitat-altering disturbances within each
basin. Variations in the frequency and magnitude of
debris-flow events may also be manifest in current
distributions of channel and valley floor morphologic
attributes (Benda, 1990; Wohl and Pearthree, 1991) and
in the size of debris-flow fans (May, 2001), providing a
means for empirical verification of such hypotheses.
2.2.2. Fire
Myriad processes affect fire behavior, resulting in
complex patterns of fire occurrence over space and
time (Agee, 1993). Efforts to mechanistically model
fire behavior require large inputs of empirical data
(Keane et al., 1996), more perhaps than are feasible for
the large-scale (McKenzie et al., 1996) and long-term
models needed to characterize fire effects on erosion
and mass-wasting regimes. Simple characterizations
of fire occurrence covering basin to regional scales and
spanning centuries can be expressed in terms of the
mean rotation interval, a distribution of fire sizes, and
initiation frequency. Numerical models incorporating
stochastic aspects of fire size and ignition can then be
used to simulate fire sequences (Agee and Flewelling,
1983). From this basis, a variety of other controlling
factors can be included in long-term models. For exam-
ple, Benda and Dunne (1997b) included increased
ignition probability following fire, Wimberly (2002,
see also Wimberly et al., 2000) included variable burn
severity, and Benda et al. (1998; see also USDA Forest
Service, 2002) included topographic controls on fire
spread. The primary constraints for use of such models
are availability of data to characterize the controlling
factors.
Estimates of fire rotation can be made from den-
drochronology of fire scars (e.g. Agee et al., 1990) and
stand-age distributions (Van Wagner, 1978), both of
which depend on the age distribution at the time of
observation. Using a fire-simulation model, Wimberly
et al. (2000) showed that stand-age distributions
can vary dramatically over time with a range of
variation that is a function of the area observed
132 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
(see also Sprugel, 1991). Stratigraphic studies (Long
et al., 1998; Millspaugh et al., 2000; Mohr et al., 2000)
provide estimates over much longer time intervals and
show temporal variability in average fire recurrence
intervals over a range of scales (Whitlock et al., this
issue).
Estimates for the size distribution of fires come
from observed historical fires (e.g. Strouss et al., 1989)
and from dendroecological reconstructions of past
fires (Teensma, 1987; Impara, 1997). Such studies
indicate positively skewed size distributions charac-
terized by many small fires and few large ones. In
long-term fire-simulation models, Benda and Dunne
(1997b) characterized fire size using a negative expo-
nential distribution and Wimberly et al. (2000) used a
size distribution characterized by the inverse of the
mean size.
2.2.3. Storm climate
The storm events that trigger erosional processes
can also be characterized using probability distribu-
tions. Benda and Dunne (1997b; see also Lancaster
et al., 2000), e.g. used independent negative exponen-
tial distributions of storm frequency and duration,
based on work by Eagleson (1972). Miller (in USDA
Forest Service, 2002) used a multiparameter distribu-
tion that accounts for correlations between frequency
and duration.
These examples are calibrated to empirical rainfall
records and can be used to generate a sequence of
events that reproduce the stochastic nature of storms
over time, but which lack spatial variability. Hydrol-
ogists must deal with issues of spatial variability as
well, as shown by a long and prolific literature on the
subject of spatial scaling for design precipitation
Fig. 8. The probability of debris-flow impacts, including scour and deposition, based on topography for Knowles Creek and Sweet Creek
basins, Coast Range, Oregon. Topographic controls on debris-flow initiation and runout length are calibrated to debris-flow events mapped by
the ODF (Robison et al., 1999) during the storm of February 1996 discussed in the text.
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 133
events (e.g. US Weather Bureau, 1957; Rodriguez-
Iturbe and Mejia, 1974; Waymire et al., 1984;
Rodriguez-Iturbe, 1986; Sivapalan and Bloschl,
1998; Seed et al., 1999 are a small sample). One of
the primary tools used to describe this concept is the
areal reduction factor (ARF), which estimates the
fractional depth of point rainfall expected when con-
sidering a basin with some finite area. These curves are
used with intensity-duration-frequency curves derived
from statistics of individual precipitation gages to
design flow structures and retention basins for catch-
ments with some area. In general, they show that the
average precipitation of a large area is less than that of
a point. This is more pronounced for peaks with short
duration because such peaks are typically associated
with storms with small spatial extent. Besides simple
spatial variability in precipitation, variability in soil
water inputs caused by varying snowmelt with eleva-
tion during rain-on-snow events would also be a strong
control on spatial coherence in climatic drivers to
geomorphically significant events.
Conceptually, introducing ARFs into time series
modeling of hillslope and channel sediment fluxes
could be accomplished fairly directly using stochastic
climate modeling. Note that depth-duration-frequency
(DDF) curves essentially quantify the degree of con-
centration of rainfall in time, and one can view the
ARFs as quantifying the degree to which the rainfall is
concentrated in space. Parametric (e.g. Hanson et al.,
1994) or resampling-based (Rajagopalan and Lall,
1999) approaches have been used to generate point
rainfall depth-duration combinations, with somewhat
better results from resampling. Recognizing that the
ARFs essentially add another dimension to IDF
curves, one could conceptualize it as a separate IDF
chart for different areas, with the changes in the depth
for each duration between charts given by the ARF.
Depth, duration, and a randomly drawn frequency
would specify an area.
There are some minor theoretical shortcomings to
this approach that require attention. Many ARFs are
fixed-area ARFs, when really what is needed is a
storm-centered ARF. Storm-centered ARFs are gen-
erally only slightly smaller (Bloschl, 1996). In addi-
tion, very little care has been taken in understanding
the seasonality of ARFs and that the ARF may change
shape with return period (Bloschl, 1996). ARF curves
depend strongly on how precipitation in one location
is correlated with precipitation in another location.
During large synoptic storms, one expects coherence
over fairly large distances, e.g. several tens of kilo-
meters. During convective precipitation events corre-
lation may only exist on the scale of about 1 km
(Bloschl, 1996). The fact that there is seasonality to
the shape of ARF curves is important when discussing
impacts of fire in forested ecosystems. Fire-induced
water repellency occurring in some vegetation and
soil types may lead to the formation of gullies for
particular rainfall intensities (Istanbulluoglu et al.,
2002). Because higher rainfall intensities are typically
associated with convective storms, spatial temporal
modeling of geomorphology in areas where this is
important would need to consider the low spatial
correlation of rainfall.
One additional challenge in applying ARFs to this
modeling approach is that when there is no spatial
correlation in rainfall, different points within a catch-
ment are essentially operating independently. Thus for
a given area, we may potentially need to simulate
more than one storm. In the simple view adopted in
earlier modeling, where the entire model domain was
subjected to the same rainfall, complete correlation
is assumed. When loss of correlation is considered
(either from increased model domain or shorter cor-
relation lengths), we need to consider the possibility of
storm cells located in more than one part of the model
domain at a time. There is no simple approach for
this problem, and a range of spatio-temporal models
for precipitation simulation have arisen to solve it
(e.g. Waymire et al., 1984; Rodriguez-Iturbe, 1986;
Seed et al., 1999; Seed, 2001). Conceptually one could
also resample radar image sequences by expanding on
existing vector resampling methods (e.g. Rajagopalan
and Lall, 1999).
Note that the spatial scale problems outlined
here are a fundamental problem with simple risk
analyses applied for the US Forest Service’s Burned
Area Emergency Rehabilitation (BAER) program (see
http://www.fs.fed.us/biology/watershed/burnareas/)
which assumes uniformity of precipitation over a fire
regardless of fire size. For large fires, it is unlikely that
a storm will cover the whole fire, but somewhat likely
that somewhere within a large fire, severe conditions
will occur over a small area.
Potentially, we can consider effects of spatial varia-
bility in snowmelt with greater ease for areas where it
134 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
is an important process in triggering large geomorphic
events. Variability in snowmelt during rain-on-snow
events appears at gross scales to be controlled by
elevation, essentially as a surrogate for temperature.
Below a given elevation, liquid precipitation com-
bined with positive sensible latent and sensible heat
to the snowpack can produce substantial soil water
inputs, whereas above that elevation precipitation falls
mostly as snow, and turbulent fluxes generally cool the
snowpack. By generating stochastic sequences that
maintain dependence structures between precipitation
and temperature (Rajagopalan and Lall, 1999), we can
estimate soil water input as a function of elevation
from a snowmelt model in areas where the process
might be important.
3. Modeling process interactions
3.1. Numerical models
As we have discussed, the factors that control
sediment fluxes exhibit variability over a range of
scales and quantification of these factors must accom-
modate variability over tens to hundreds of kilometers
and hundreds, perhaps thousands, of years. Interac-
tions that drive sediment inputs may be impossible
to discern using measured spatial and temporal
sequences at smaller scales (Kirchner et al., 2001),
yet knowledge of factors controlling their frequency
and magnitude, the spatial extent of affected stream
channels, and spatial synchrony between basins is
crucial to anticipating affects of land management
and climate change on aquatic ecosystems (Reeves
et al., 1995, Dunham et al., this issue).
To skirt the limitations imposed by observations
that span only a few decades, Benda and Dunne
(1997a,b) used numerical models to examine process
interactions over millennial time scales in a 200 km2
basin in western Oregon. They showed how super-
imposed storm and fire sequences, acting over hetero-
geneous topography, drive episodes of sediment
delivery to a channel system. Use of a numerical
model allowed them to generate event sequences over
time and space scales large enough to reveal formation
of distinct patterns in sediment transport and storage
through a channel network and to show how such
patterns emerge from the interaction of fires, storms,
topography, and channel-network geometry over an
entire basin. The existence and source of such patterns
must still be established through field studies, but their
prediction shows how process interactions over large
scales may be manifest and can guide field investiga-
tions to test these ideas.
Benda and Dunne (1997a,b) illustrated the use of a
‘‘top-down’’ or ‘‘hierarchical’’ modeling approach
(Murray, 2002), in which processes are described at
the scale of interest (Werner, 1999). In their case, that
scale was that of a channel reach—hundreds of
meters—over annual time steps. They generalized pro-
cesses of sediment flux and fluvial transport over these
scales. This strategy seeks process descriptions at scales
relevant to the system at hand, e.g. a river network, and
requires parameterizations that reduce degrees of free-
dom to those active at that scale. This is in contrast to a
reductionist strategy, which seeks to describe large-
scale systems from the ‘‘bottom-up’’, starting from the
underlying processes evident at smaller scales, e.g.
fluvial transport of sediment grains. Both approaches
add to our understanding of natural systems, but hier-
archical models are needed to explore interactions at
scales pertinent to the frequency, magnitude, spatial
extent, and synchrony of channel disturbance.
Hierarchical models have a variety of uses. They
provide a means of visualizing concepts over basin
scales. Benda et al. (1998) used model results to
illustrate effects of basin size and position in a channel
network on sediment yields and probability distribu-
tions of sediment storage in channel reaches. TheUSDA
Forest Service (2002) used model animations to show
how fires, storms, landsliding, and fluvial transport can
interact to produce variability in large woody debris
and sediment storage through a river network.
These models can be used to explore the conse-
quences of process interactions. Benda and Dunne
(1997a) predicted that the overlapping effects of fires
and storms drive episodic influxes of sediment to
channels that initiate pulses of sediment transport
through the network. In contrast, Lancaster et al.
(2000) predicted that effects of woody debris on
debris-flow runout distance cause low-order basins
to act as long-term sediment stores that gradually
meter sediment out to larger channels, thereby diffus-
ing the capacity for episodes of landsliding to generate
pulses of fluvial transport through the high-order net-
work. Differences in model predictions highlight the
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 135
potential role of specific factors and motivate field
measurements to test these hypotheses.
As experience with these types of models grows,
their utility will expand. They will be used as explora-
tory tools to evaluate the influence that differences in
topography or changes in climate or fire regime might
have on sediment fluxes and associated channel dis-
turbances. They will be used to evaluate the time
scales of response and recovery and the role of history
in setting channel and basin responses to storm and fire
events. Importantly, they will be used in scenario
testing to evaluate the likely consequences of different
management strategies (Dunne et al., 2001).
3.2. Data requirements
Two general types of information are required to
build numerical models: (1) data to characterize the
driving events, e.g. fires and storms, and (2) data to
characterize controls on consequent erosional events.
The models referenced here lack several important
factors: characterization of fire distributions from
stand-age mapping as used by Benda et al. (1998)
misses potentially important aspects of fire intensity;
use of spatially uniform storm events misses important
aspects of spatial variability in storm intensity; and
currently available digital elevation data cannot
resolve small-scale topographic controls on gully
initiation and landsliding. However, progress is being
made on all fronts. Aerial mapping of fire intensity
provides information on fire size distributions and
field mapping of associated effects (e.g. changes in
soil infiltration capacity) links fire intensity to con-
sequent impacts. Doppler radar may provide a means
of characterizing the spatial variability of rainfall
intensity associated with single storms and provide
information to estimate the size and frequency dis-
tribution of storm events. New methods for remotely
sensed elevation measurements, such as laser altime-
try (LIDAR), can provide high-resolution topographic
data. The next challenge may be to find the resources
to use available data.
Currently, basin-scale models can produce tempo-
rally and spatially distributed estimates of sediment
flux. Process-based models to translate estimates of
sediment flux and storage to temporal and spatial
predictions of biologically pertinent channel charac-
teristics, such as substrate texture (Dietrich et al., 1989;
Buffington and Montgomery, 1999; Lisle et al., 2000)
or pool extent (USDA Forest Service, 2002) must still
be incorporated. Data to characterize the consequences
to channels of changes in sediment volume and to
characterize biological responses to those changes are
essential for using numerical models to interpret and
infer the consequences of past and future changes in
fire regime, climate, and land management.
4. Conclusions
Sediment fluxes and consequent stream-channel
disturbances are driven by process interactions over
time periods spanning centuries and spatial scales
spanning entire river basins. Methods for data collec-
tion and analyses over these scales are being actively
developed, spurred by the realization that land-man-
agement practices, coupled with changes in climate,
can dramatically influence rates of sediment delivery
to stream channels. Stratographic and sedimentologi-
cal studies (e.g. Meyer and Pierce, this issue) are
greatly expanding records of past sediment-flux events
and point to strong climatic controls and large varia-
tions across regions. To anticipate the consequences of
land management, fire suppression, and climate
change, we require quantitative characterizations of
the processes that drive sediment fluxes and of the
factors that control the spatial and temporal patterns of
sediment delivery to stream channels. The stochastic
and heterogeneous nature of these processes leads us
to probabilistic descriptions and to development of
analytical methods that use probabilistic parameter-
izations. Predictive tools must also deal with the
effects of past events. These requirements have lead
to development of numerical models that can simulate
process interactions over large temporal and spatial
scales. As experience with these types of models
accumulates, they have the potential to provide a
useful analysis tool, offering a spatial and temporal
context for interpretation of field observations and
predictions that can be tested with field measurements.
Acknowledgements
The landslide density and topographic control on
debris-flow probability analyses reported on here were
136 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
done with support and assistance from the Coastal
Landscape Analysis and Modeling study through the
US Forest Service, Pacific Northwest Research Sta-
tion, Forest Sciences Laboratory, in Corvallis, Oregon.
Jim Paul and Jason Hinkle with the Oregon Depart-
ment of Forestry were also very helpful in providing
data from the 1996 storm study. We thank Christine
May, who provided data from her thesis work and
prompted many helpful discussions. Thanks to Kelly
Christiansen and Dave Nagel for assisting with
graphics production and to Jack King for aerial photo-
graph interpretation and mapping. We greatly appreci-
ate reviews by David Tarboton, John Buffington, and
an anonymous reviewer, which served to improve an
earlier version of this paper.
References
Agee, J.K., 1993. Fire Ecology of Pacific Northwest Forests. Island
Press, Washington, DC.
Agee, J.K., Flewelling, R. (Eds.), 1983. A Fire Cycle Model Based
on Climate for the Olympic Mountains, Washington. In:
Proceedings of Seventh Conference on Fire and Forest
Meteorology. American Meteorology Society, Boston, 7,
pp. 32–37.
Agee, J.K., Finney, M., de Gouvenain, R., 1990. Forest fire history
of Desolation Peak, Washington. Can. J. For. Res. 20, 350–356.
Bear, J., 1972. Dynamics of Fluids in Porous Media. Elsevier, New
York.
Benda, L.E., 1990. The influence of debris flows on channels and
valley floors in the Oregon Coast Range, USA. Earth Surf.
Process. Landforms 15, 457–466.
Benda, L.E., Cundy, T.W., 1990. Predicting deposition of
debris flows in mountain channels. Can. Geotechnol. J. 27,
409–417.
Benda, L.E., Dunne, T., 1997a. Stochastic forcing of sediment
routing and storage in channel networks. Water Resour. Res.
33, 2865–2880.
Benda, L.E., Dunne, T., 1997b. Stochastic forcing of sediment
supply to channel networks from landsliding and debris flow.
Water Resour. Res. 33, 2849–2863.
Benda, L.E., Miller, D.J., Dunne, T., Reeves, G.H., Agee, J.K.,
1998. Dynamic landscape systems. In: Naiman, R.J., Bilby,
R.E. (Eds.), River Ecology and Management. Springer, New
York, pp. 261–288.
Bloschl, G., 1996. Scale and Scaling in Hydrology. Weiner
Mitteilungen Wasser Abwasser Gewasser, Wien.
Brardinoni, F., 2001. Identification of natural and logging-related
landslides in the Capilano River Basin (Coastal British
Columbia): a comparison between remotely sensed survey
and field survey. Masters Thesis. The University of British
Columbia, Vancouver, BC.
Buffington, J.M., Montgomery, D.R., 1999. Effects of sediment
supply on surface textures of gravel-bed rivers. Water Resour.
Res. 35, 3523–3530.
Bush, G., McConnel, C., Cloyd, C., Musser, K., Metzger, B.,
Plumley, H., 1997. Assessment of the effects of the 1996 flood
on the Siuslaw National Forest. USDA Forest Service, Siuslaw
National Forest, Corvallis, OR.
Caine, N., 1980. The rainfall intensity-duration control of shallow
landslides and debris flows. Geografiska Annaler. A 62, 23–27.
Cannon, S.H., 2001. Debris-flow generation from recently burned
watersheds. Environ. Eng. Geosci. 7, 321–341.
Cenderelli, D.A., Kite, J.S., 1998. Geomorphic effects of large
debris flows on channel morphology at North Fork Mountain,
Eastern West Virginia, USA. Earth Surf. Process. Landforms
23, 1–19.
Chernick, M.R., 1999. Bootstrap Methods: A Practitioner’s Guide.
Wiley, New York.
Coates, R., Collins, L., 1984. Streamside landsliding and channel
change in a suburban forested watershed: effects of an extreme
event. In: Proceedings of the Symposium on Effects of Forest
Land Use on Erosion and Slope Stability. East-West Center,
Honolulu, pp. 155–164.
Coho, C., Burges, S.J., 1993. Dam-break floods in low-order
mountain channels of the Pacific Northwest. Water Resources
Series, Technical Report No. 138. Department of Civil
Engineering, University of Washington, Seattle, WA.
Dietrich, W.E., Dunne, T., 1978. Sediment budget for a small
catchment in mountainous terrain. Z. Geomorph. Suppl. Bd. 29,
191–206.
Dietrich, W.E., Kirchner, J.W., Ikeda, H., Iseya, F., 1989. Sediment
supply and the development of the coarse surface layer in
gravel-bedded rivers. Nature 340, 215–217.
Dietrich, W.E., Wilson, C.J., Montgomery, D.R., McKean, J.,
Bauer, R., 1992. Erosion thresholds and land surface morphol-
ogy. Geology 20, 675–679.
Dietrich, W.E., Bellugi, D., de Asua, R.R., 2001. Validation of the
shallow landslide model, SHALSTAB, for forest management.
In: Wigmosta, M.S., Burges, S.J. (Eds.), Land Use and
Watersheds. American Geophysical Union, Washington, DC,
pp. 195–227.
Dragovich, J.D., Brunengo, M.J., Gerstel, W.J., 1993a. Landslide
inventory and analysis of the Tilton River—mineral Creek area,
Lewis county, Washington. Part 1. Terrain and Geologic
Factors. Wash. Geol. 21, 9–18.
Dragovich, J.D., Brunengo, M.J., Gerstel, W.J., 1993b. Landslide
inventory and analysis of the Tilton River—mineral Creek area,
Lewis county, Washington. Part 2. Soils, harvest age, and
conclusions. Wash. Geol. 21, 18–30.
Dunham, J.B., Rieman, B.E., 1999. Metapopulation structure of
bull trout: influences of physical, biotic, and geometrical
landscape characteristics. Ecol. Appl. 9, 642–655.
Dunne, T., 1998. Critical data requirements for prediction of
erosion and sedimentation in mountain drainage basins. J. Am.
Water Resour. Assoc. 34, 795–808.
Dunne, T., Agee, J.K., Beissinger, S., Dietrich, W.E., Gray, D.,
Power, M., Resh, V., Rodrigues, K., 2001. A scientific basis for
the prediction of cumulative watershed effects. University of
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 137
California Wildland Resource Center Report No. 46. University
of California Wildland Resource Center, Berkeley, CA.
Eagleson, P.S., 1972. Dynamics of flood frequency. Water Resour.
Res. 8, 878–898.
Everest, F.H., Meehan, W.R., 1981. Forest management and
anadromous fish habitat productivity. Trans. North Am. Wildl.
Nat. Resour. Conf. 46, 521–530.
Everest, F.H., Beschta, R.L., Scrivener, J.C., Koski, K.V., Sedell,
J.R., Cederholm, C.J., 1987. Fine sediment and salmonid
production: A Paradox. In: Salo, E.O., Cundy, T.W. (Eds.),
Streamside Management: Forestry and Fishery Interactions.
College of Forest Resources, University of Washington, Seattle,
WA, USA, pp. 98–142.
Grant, G.G., Swanson, F.J., 1995. Morphology and processes of
valley floors in mountain streams, Western Cascades, Oregon.
In: Costa, J.E., Miller, A.J., Potter, K.W., Wilcock, P.R. (Eds.),
Natural and Anthropogenic Influences in Fluvial Geomorphol-
ogy, Geophysical Monograph 89. American Geophysical
Union, Washington, DC, pp. 83–101.
Grant, G.E., Swanson, F., Wolman, M.G., 1990. Pattern and origin
of stepped-bed morphology in high-gradient streams, Western
Cascades, Oregon. GSA Bull. 102, 340–352.
Gresswell, R.E., 1999. Fire and aquatic ecosystems in forested
biomes of North America. Trans. Am. Fish. Soc. 128, 193–221.
Hack, J.T., Goodlett, J.C., 1960. Geomorphology and Forest
ecology of a Mountain Region in the Central Appalachians.
US Geological Survey Professional Paper 347, 66 pp.
Hall, D.E., Long, M.T., Remboldt, M.D. (Eds.), 1994. Slope
Stability Reference Guide for National Forests in the United
States. United States Department of Agriculture, Forest
Service, Washington, DC.
Hammond, C., Hall, D., Miller, S., Swetik, P., 1992. Level 1
Stability Analysis (LISA) Documentation for Version 2.0.
General Technical Report INT-285. United States Department
of Agriculture, Forest Service, Intermountain Research Station,
Ogden, Utah.
Hanson, C.L., Cumming, K.A., Woolhiser, D.A., Richardson, C.W.,
1994. Microcomputer Program for Daily Weather Simulations
in the Contiguous United States. ARS-114. Agricultural
Research Service, US Department of Agriculture, Washington,
DC.
Harvey, A.M., 1991. The influence of sediment supply on the
channel morphology of upland streams: Howgill Fells, North-
west England. Earth Surf. Process. Landforms 16, 675–684.
Hogan, D.L., Bird, S.A., Rice, S., 1998. Stream channel
morphology and recovery processes. In: Hogan, D.L.,
Tschaplinski, P.J., Chatwin, S. (Eds.), Carnation Creek and
Queen Charlotte Islands Fish/Forestry Workshop: Applying 20
Years of Coast Research to Management Solutions. Land
Management Handbook 41. Crown Publications, Victoria, BC,
pp. 77–96.
Hovius, N., Stark, C.P., Allen, P.A., 1997. Sediment flux from
a mountain belt derived by landslide mapping. Geology 25,
231–234.
Impara, P.C., 1997. Spatial and temporal patterns of fire in the
forests of the Central Oregon Coast Range. Ph.D. Oregon State
University, Corvallis, OR.
Istanbulluoglu, E., Tarboton, D.G., Pack, R.T., Luce, C.H., 2002. A
probabilistic approach for channel initiation. Water Resour.
Res. 38, 1325.
Iverson, R.M., 2000. Landslide triggering by rain infiltration. Water
Resour. Res. 36, 1897–1910.
Iverson, R.M., Schilling, S.P., Vallance, J.W., 1998. Objective
delineation of lahar-inundation hazard zones. Geol. Soc. Am.
Bull. 110, 972–984.
Keane, R.E., Ryan, K., Running, S.W., 1996. FIRE-BGC—a
mechanistic ecological process model for simulating fire
succession on coniferous forest landscapes of the Northern
Rocky Mountains. INT-RP-484. USDA Forest Service, Fort
Collins, CO.
Kirchner, J.W., Finkel, R.C., Riebe, C.S., Granger, D.E., Clayton,
J.L., King, J.G., Megahan, W.F., 2001. Mountain erosion over
10 year, 10 k.y., and 10 m.y. time scales. Geology 29, 591–594.
Lancaster, S.T., Hayes, S.K., Grant, G.E., 2000. Modeling sediment
and wood storage and dynamics in small mountainous water-
sheds. In: Dorava, J.M., Montgomery, D.R., Palcsak, B.B.,
Fitzpatrick, F.A. (Ed.), Geomorphic Processes and Riverine
Habitat, vol. 4. American Geophysical Union, Washington, DC,
pp. 85–102.
Larsen, M.C., Simon, A., 1993. A rainfall intensity-duration
threshold for landslides in a humid-tropical environment,
Puerto Rico. Geografiska Annaler. 75, 13–23.
Lisle, T.E., Nelson, J.M., Pitlick, J., Madej, M.A., Barkett, B.L.,
2000. Variability of bed mobility in natural, gravel-bed
channels and adjustment to sediment load at local and reach
scales. Water Resour. Res. 36, 3743–3755.
Long, C.J., Whitlock, C., Bartlein, P.J., Millspaugh, S.H., 1998. A
9000-year fire history from the Oregon coast range, based on a
high-resolution charcoal study. Can. J. For. Res. 28, 774–787.
Madej, M.A., 1990. Changes in channel-stored sediment, redwood
creek, Northwestern California, 1947–1980. In: Nolan, K.M.,
Kelsey, H.M., Marron, D.C. (Eds.), Geomorphic Processes and
Aquatic Habitat in the Redwood Creek Basin, Northwestern
California. US Geological Survey Professional Paper 1454,
Washington, DC.
Madej, M.A., Ozaki, V., 1996. Channel response to sediment wave
propagation and movement, Redwood Creek, California, USA.
Earth Surf. Process. Landforms 21, 911–927.
May, C.L., 1998. Debris flow characteristics associated with forest
practices in the Central Oregon Coast range. Masters. Oregon
State University, Corvallis, OR.
May, C.L., 2001. Spatial and temporal dynamics of sediment and
wood in headwater streams in the Central Oregon Coast range.
Ph.D. Oregon State University, Corvallis, OR.
May, C.L., 2002. Debris flows through different forest age classes
in the central Oregon Coast Range. J. Am. Water Resour.
Assoc. 38, 1–17.
McKenzie, D., Peterson, D.L., Alvarado, E., 1996. Extrapolation
problems in modeling fire effects at large spatial scales: a
review. Int. J. Wildl. Fire 6, 165–176.
Meyer, G.A., Wells, S.G., Jull, A.J.T., 1995. Fire and alluvial
chronology in Yellowstone National Park: climatic and intrinsic
controls on Holocene geomorphic processes. Geol. Soc. Am.
Bull. 107, 1211–1230.
138 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140
Meyer, G.A., Pierce, J.L., Wood, S.H., Jull, A.J.T., 2001. Fire,
storms, and erosional events in the Idaho batholith. Hydrol.
Process. 15, 3025–3038.
Middleton, G.V., Wilcock, P.R., 1994. Mechanics in the Earth
and Environmental Sciences. Cambridge University Press,
Cambridge.
Miller, D.J., Benda, L.E., 2000. Effects of punctuated sediment
supply on valley-floor landforms and sediment transport. Geol.
Soc. Am. Bull. 112, 1814–1824.
Miller, D.J., Burnett, K.M., Christiansen, K., Clarke, S., 2002. A
simple model for regional prediction of debris flow impacts. In:
http://www.fsl.orst.edu/clams/, Projects, Broad-scale Models of
Landslides and Debris Flows.
Millspaugh, S.H., Whitlock, C., Bartlein, P.J., 2000. Variations in
fire frequency and climate over the past 17,000 year in central
Yellowstone National Park. Geology 28, 211–214.
Mohr, J.A., Whitlock, C., Skinner, C.N., 2000. Postglacial
vegetation and fire history, eastern Klamath mountains,
California, USA. The Holocene 10, 587–601.
Montgomery, D.R., Buffington, J.M., 1998. Channel processes,
classification, and response. In: Naiman, R.J., Bilby, R.E. (Eds.),
River Ecology and Management. Springer, New York, pp. 13–42.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based
model for the topographic control on shallow landsliding.
Water Resour. Res. 30, 1153–1171.
Montgomery, D.R., Schmidt, K.M., Greenberg, H.M., Dietrich,
W.E., 2000. Forest clearing and regional landsliding. Geology
28, 311–314.
Murray, B.A., 2002. Seeking explanation affects numerical modeling
strategies. EOS, Trans. Am. Geophys. Union 83, 418–419.
Nakamura, F., Maita, H., Araya, T., 1995. Sediment routing
analyses based on chronological changes in hillslope and
riverbed morphologies. Earth Surf. Process. Landforms 20,
333–346.
Nolan, K.M., Marron, D.C., 1988. Stream-channel response to the
storm in the Santa Cruz mountains. In: Ellen, S.D., Wieczorek,
G.F. (Eds.), Landslides, Floods, and Marine Effects of the
Storm of January 3–5, 1982, in the San Francisco Bay Region,
California. US Geological Survey Professional Paper 1434. US
Geological Survey, Washington, DC.
Nolan, K.M., Marron, D.C., 1990. History, causes, and significance
of changes in the channel geometry of Redwood Creek,
northwestern California, 1926–1982. In: Nolan, K.M., Kelsey,
H.M., Marron, D.C. (Eds.), Geomorphic Processes and Aquatic
Habitat in the Redwood Creek Basin, Northwestern California.
US Geological Survey Professional Paper 1454. US Geological
Survey, Washington, DC.
Ohmann, J.L., Gregory, M.J., 2002. Predictive mapping of forest
composition and structure with direct gradient analysis and
nearest neighbor imputation in coastal Oregon, USA. Can. J.
For. Res. 32, 725–741.
O’Laughlin, E.M., 1986. Prediction of surface saturation zones in
natural catchments by topographic analysis. Water Resour. Res.
22, 794–804.
Oregan State University, 1996. Summary of precipitation for
central Oregon Coast Range, October 1995 to April 1996.
COPE Report 9 (4). Oregon State University, Corvallis.
Pabst, R.J., Spies, T.A., 2001. Ten years of vegetation succession
on a debris-flow deposit in Oregon. J. Am. Water Resour.
Assoc. 37, 1693–1708.
Pyles, M.R., Froehlich, H.A., 1987. Discussion of ‘‘Rates of
Landsliding as Impacted by Timber Management Activities in
Northwestern California’’. Bull. Assoc. Eng. Geol. 24, 425–
431.
Rajagopalan, B., Lall, U., 1999. A k-nearest-neighbor simulator for
daily precipitation and other weather variables. Water Resour.
Res. 35, 3089–3101.
Reeves, G.H., Benda, L.E., Burnett, K.M., Bisson, P.A., Sedell,
J.R., 1995. A disturbance-based ecosystem approach to
maintaining and restoring freshwater habitats of evolutionarily
significant units of anadromous salmonids in the Pacific
Northwest. In: Nielson, J.L., Powers, D.A. (Eds.) Evolution
and the Aquatic Ecosystem: Defining Unique Units in
Population Conservation, American Fisheries Society Sympo-
sium 17. American Fisheries Society, Bethesda, MD, USA,
pp. 334–349.
Reneau, S.L., Dietrich, W.E., Donahue, D.J., Jull, A.J.T., Rubin,
M., 1990. Late quaternary history of colluvial deposition and
erosion in hollows, central California Coast Ranges. Geol. Soc.
Am. Bull. 102, 969–982.
Robison, G.E., Mills, K.A., Paul, J., Dent, L., Skaugset, A., 1999.
Storm Impacts and Landslides of 1996: Final Report 4. Oregon
Department of Forestry.
Rodriguez-Iturbe, I., 1986. Scale of fluctuation of rainfall models.
Water Resour. Res. 22, 15S–37S.
Rodriguez-Iturbe, I., Mejia, J.M., 1974. On the transformation
from point rainfall to areal rainfall. Water Resour. Res. 10,
729–735.
Schmidt, K.M., 1999. Root strength, colluvial soil depth, and
colluvial transport on landslide-prone hillslopes. Ph.D. Dis-
sertation. University of Washington, Seattle.
Seed, A.W., 2001. Theory and measurements of the distribution
of rainfall in space and time. In: Proceedings of the
MODSIM 2001, International Congress on Modelling
and Simulation, Canberra, Australia, December 10–13.
Modelling and Simulation Society of Australia and New
Zealand.
Seed, A.W., Srikanthan, R., Menabde, M., 1999. A space and
time model for design storm rainfall. J. Geophys. Res. 104,
31623–31630.
Sidle, R.C., Pearce, A.J., O’Loughlin, C.L., 1985. Hillslope
Stability and Land Use. American Geophysical Union,
Washington, DC.
Sivapalan, M., Bloschl, G., 1998. Transformation of point rainfall
to areal rainfall: intensity-duration-frequency curves. J. Hydrol.
204, 150–167.
Sprugel, D.G., 1991. Disturbance, equilibrium, and environmental
variability: What is ‘natural’ vegetation in a changing
environment? Biol. Conserv. 58, 1–18.
Strouss, D., Bednar, L., Mees, R., 1989. Do one percent of forest
fires cause ninety-nine percent of the damage? For. Sci. 35,
319–328.
Swanson, F.J., 1981. Fire and geomorphic processes. In: Fire
Regimes and Ecosystem Properties, General Technical Report
D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140 139
W0-26. US Department of Agriculture Forest Service,
Washington, DC, pp. 401–420.
Swanson, F.J., Kratz, T.K., Caine, N., Woodmansee, R.G., 1988.
Landform effects on ecosystem patterns and processes.
BioScience 38, 92–98.
Taylor, G., 1997a. Causes of the flood and a comparison to other
climate events. In: Ruff, J.D. (Eds.), The Pacific Northwest
Floods of February 6–11, 1996: Proceedings of the Pacific
Northwest Water Issues Conference. American Institute of
Hydrology, St. Paul, Minnesota, USA, pp. 3–7.
Taylor, G.H., 1997b. The Great Flood of 1996, Oregon Climate
Service.
Teensma, P.D.A., 1987. Fire history and fire regimes of the central
western Cascades of Oregon. Ph.D. University of Oregon,
Eugene, OR.
US Weather Bureau, 1957. Rainfall intensity-frequency regime 1.
The Ohio valley 2. Southeastern United States. Technical
Paper No. 29. US Department of Commerce, Washington,
DC.
USDA Forest Service, 2002. Landscape dynamics and forest
management. Gen. Tech. Rep. RMRS-GTR-101CD. US
Department of Agriculture, Forest Service, Rocky Mountain
Research Station, Fort Collins, CO.
Van Wagner, C.E., 1978. Age-class distribution and the forest fire
cycle. Can. J. For. Res. 8, 220–227.
Waymire, E., Gupta, V.K., Rodriguez-Iturbe, I., 1984. A spectral
theory of rainfall intensity at the meso-‘‘Beta’’ scale. Water
Resour. Res. 20, 1453–1465.
Werner, B.T., 1999. Complexity in natural landform patterns.
Science 284, 102–104.
Wimberly, M.C., 2002. Spatial simulation of historical landscape
patterns in coastal forests of the Pacific Northwest. Can. J. For.
Res. 32, 1316–1328.
Wimberly, M.C., Spies, T.A., Long, C.J., Whitlock, C., 2000.
Simulating historical variability in the amount of old forests in
the Oregon Coast Range. Conserv. Biol. 14, 167–180.
Wohl, E.E., Pearthree, P.P., 1991. Debris flows as geomorphic
agents in the Huachuca Mountains of southeastern Arizona.
Geomorphology 4, 273–292.
Zevenbergen, L.W., Thorne, C.R., 1987. Quantitative analysis of
land surface topography. Earth Surf. Process. Landforms 12,
47–56.
140 D. Miller et al. / Forest Ecology and Management 178 (2003) 121–140